Innovation Session: Microsoft 365 Copilot: AI Built For Work | BRK1710

Introduction to Microsoft 365 Copilot00:17

Slide for Introduction to Microsoft 365 Copilot

Nicole Herskowitz, Corporate Vice President, opens the presentation by emphasizing the team's dedication to building a product that users love and consider indispensable. She highlights the focus on creating a tool that integrates seamlessly into daily workflows.

Rapid Innovation and Feature Releases00:31

Slide for Rapid Innovation and Feature Releases

Microsoft has shipped over **400 new features** in response to user feedback, demonstrating a commitment to speed and innovation. Key advancements include: * **GPT-4o Integration:** Shipped on the same day it was launched. * **Researcher with Computer Use:** Allows users to tell Copilot what to do, and it executes the task. * **Sora Integration:** Bringing video generation capabilities to the Copilot Create experience. * **Model Choice:** Incorporating **Anthropic models** to power new innovations.

Enterprise Adoption and Core Fundamentals02:03

Slide for Enterprise Adoption and Core Fundamentals

While innovating rapidly, Microsoft maintains a focus on core fundamentals: high performance, reliability, and security. The impact of this strategy is reflected in market adoption, with **90% of the Fortune 500** currently using Microsoft 365 Copilot.

Customer Success Stories02:08

Slide for Customer Success Stories

Major corporations are not just adopting Copilot but are reshaping their operating models with it: * **Barclays:** Making banking more personal and intuitive for customers. * **DuPont:** Helping scientists accelerate the process from hypothesis to breakthrough. * **Levi's:** Assisting retail employees in better serving their customers.

AI Built for Work02:37

Slide for AI Built for Work

The presentation introduces the concept of "AI Built for Work," a unified experience that combines AI and agents. This approach is characterized by: * **Best-in-class Models:** Powered by top-tier AI models. * **App Integration:** Embedded directly into the apps employees use daily. * **Security:** Secure at the core within the Microsoft 365 environment. * **Business Tuning:** Taking the latest innovations and fine-tuning them specifically for business results.

Introduction to Work IQ02:34

Slide for Introduction to Work IQ

### The Intelligence Layer **Work IQ** is introduced as the intelligence layer that powers Copilot. It is designed to understand: - You - Your job - Your company It consists of three core pillars: **Data**, **Memory**, and **Inference**.

The Data Component02:43

Slide for The Data Component

### Understanding Work Context The **Data** component allows Copilot to know the user's work context, including: - Important meetings - Active files - Unread emails - Professional relationships (who you work for and with)

The Memory Component02:59

Slide for The Memory Component

### Learning Work Patterns The **Memory** component enables Copilot to learn *how* a user works. This includes: - **Style & Preferences:** Habits and personal style. - **Workflows:** How specific tasks are handled, such as customer escalations, RFP preparations, or annual reviews.

The Inference Component03:19

Slide for The Inference Component

### Synthesizing Intelligence **Inference** is where Copilot synthesizes information to: - Make connections - Unlock insights - Predict the **next best action** for the user.

Integration and Feedback Loop04:27

Slide for Integration and Feedback Loop

### Built into Microsoft 365 Work IQ is integrated directly into Microsoft 365 apps, creating a **feedback loop**. As it picks up more signals from app usage, it delivers smarter and more personalized experiences. Crucially, all data remains secure within the user's tenant.

Custom Agents04:37

Slide for Custom Agents

### Tailored Solutions Users can leverage Work IQ to build **custom agents**. These agents can be tuned to specific unique workflows and business needs.

Work IQ vs. Connectors04:46

Slide for Work IQ vs. Connectors

### The Limitation of Connectors Standard AI connectors often pull **fragments of data** without understanding context or relationships. This can lead to: - Missed critical information - Incomplete answers - Hallucinations In contrast, Work IQ operates across all data in real-time with full context.

Security and Compliance06:07

Slide for Security and Compliance

### Enterprise-Grade Security Connectors may not respect organizational controls, introducing risk. **Work IQ** is built on the trusted Microsoft 365 environment, honoring all policies and permissions to protect the company's knowledge assets.

Introduction to Work IQ04:56

Slide for Introduction to Work IQ

The presentation begins by introducing the concept of **Work IQ**. The speaker emphasizes that true AI integration requires respecting all user permissions and controls while understanding the user, their job, and the company context.

Three Patterns of Work05:02

Slide for Three Patterns of Work

The speaker outlines three new patterns of work unlocked by Copilot and agents: * **Pattern 1:** AI assistants * **Pattern 2:** AI teammates * **Pattern 3:** AI operators This session focuses on exploring these patterns, starting with AI as an assistant.

Pattern 1: AI Assistants05:21

Slide for Pattern 1: AI Assistants

The speaker defines the role of an AI assistant. Unlike simple prompt-response interactions, a true AI assistant for work must: > "Understand your context, take on tasks, and keep up as your work continuously evolves." With Work IQ, Copilot is positioned as this type of comprehensive assistant.

Demo: Copilot Contextual Reasoning06:39

Slide for Demo: Copilot Contextual Reasoning

A demonstration shows Copilot preparing for a recurring meeting with a customer, **Contoso**. The user prompts Copilot to look back at previous meetings to find open and resolved issues. **Key Capabilities Demonstrated:** * **Reasoning Model:** Copilot pieces together information not just from transcripts, but also from chat logs and content shared on screen. * **Output:** It generates a structured list of open issues (e.g., "Deliver the Project Merlin proposal") and resolved/agreed items, providing deep context for the upcoming meeting.

Comparison with ChatGPT08:21

Slide for Comparison with ChatGPT

The speaker compares the Copilot experience with **ChatGPT** using enabled connectors for Outlook, SharePoint, and Teams. When asked the exact same query regarding the Contoso meeting: * ChatGPT attempts to process the request but lacks the deep contextual understanding of the specific meeting data graph. * **Result:** It returns a response stating, "No resolved issues found" and "No open issues found," highlighting the difference in efficacy provided by Work IQ's deep integration.

ChatGPT's Limitation with Meeting Data07:24

Slide for ChatGPT's Limitation with Meeting Data

The speaker highlights a critical gap in ChatGPT's capabilities within an enterprise context. > "ChatGPT does not take into account any Teams meetings data." * **Outlook vs. Teams**: ChatGPT only retrieves data from Outlook meeting events. * **Missing Context**: It fails to access the vast amount of data generated in Teams meetings, which are essential for complete business insights.

Copilot: Retrieving Cross-App Action Items07:44

Slide for Copilot: Retrieving Cross-App Action Items

The demonstration shifts to Microsoft 365 Copilot to handle a complex query involving organizational structure and multiple data sources. ### The Prompt The user asks Copilot to summarize action items from the sales team reporting to a specific manager (Amber), looking across Teams chats, meetings, and emails. ### The Result * **Reasoning Agent**: Copilot utilizes a reasoning agent to sift through organizational data. * **Contextual Awareness**: It correctly identifies team members (e.g., Renata Hall, Dakota Sanchez) who report to Amber. * **Structured Output**: It compiles a clear checklist of action items with specific due dates derived from the retrieved content.

ChatGPT: Failure to Retrieve Contextual Data08:51

Slide for ChatGPT: Failure to Retrieve Contextual Data

The same prompt used in Copilot is run in ChatGPT to compare performance. * **Lack of Graph Access**: ChatGPT fails to identify the specific sales team or the reporting structure related to "Amber." * **Irrelevant Results**: Instead of retrieving the requested sales actions, it pulls random, unrelated tasks assigned to the user in Teams. * **Conclusion**: The tool lacks the necessary connection to the internal organizational graph to provide valuable, actionable content for this specific request.

Enterprise Security: Project Merlin09:27

Slide for Enterprise Security: Project Merlin

The final segment introduces the concept of data security and sensitivity labels. ### Project Merlin * **Confidentiality**: The speaker opens a document for "Project Merlin," a secret pilot project. * **Sensitivity Labels**: The document is encrypted and tagged with sensitivity labels, ensuring that the content cannot be shared broadly outside the organization. * **Enterprise Requirement**: This highlights the necessity for AI tools to respect data governance and security protocols when summarizing or handling sensitive internal information.

Microsoft 365 Copilot: Secure Data Access09:52

Slide for Microsoft 365 Copilot: Secure Data Access

The demonstration begins with **Microsoft 365 Copilot** querying a specific internal document, the *Project Merlin Product Brief*. * **Contextual Awareness**: Copilot successfully summarizes the document because it recognizes the user's specific access permissions. * **Security**: The interface displays a "Highly Confidential" indicator, confirming that the data is encrypted and respected within the organization's security boundary. If an unauthorized employee attempted this, they would not receive a result.

Comparison with External AI (ChatGPT)10:39

Slide for Comparison with External AI (ChatGPT)

The speaker contrasts the previous demo by running the exact same prompt in **ChatGPT**. > "It's saying there's no document with that name or content." * **Limitation**: External tools lack access to the internal corporate graph and file permissions. * **Security Risk**: ChatGPT suggests uploading the document directly. The speaker warns this could lead to "bad actions," such as employees removing encryption and sensitivity labels to make data accessible to external AI.

Work IQ12:01

Slide for Work IQ

The concept of **Work IQ** is introduced as a differentiator for Microsoft 365 Copilot. * **Holistic Understanding**: Unlike other AI tools that use simple connectors, Copilot understands the "whole picture at work." * **Integration**: It leverages the secure graph of data, permissions, and activity unique to the organization.

Agent Mode and "Vibe Working"12:11

Slide for Agent Mode and "Vibe Working"

The lecture shifts to the future of information work, drawing an analogy to software development trends like "vibe coding." * **Vibe Working**: A new workflow where users describe an outcome or intent rather than dictating every step. * **Agent Mode**: A feature allowing users to steer Copilot iteratively, turning ideas into world-class content through collaboration rather than simple command-and-control.

Introduction: Sumit Chauhan13:27

Slide for Introduction: Sumit Chauhan

The speaker introduces **Sumit Chauhan**, Corporate Vice President of the Microsoft Office Product Group, to provide a live demonstration of Agent Mode and the new "vibe working" capabilities in action.

Introduction and Mission12:34

Slide for Introduction and Mission

Sumit Chauhan, Corporate Vice President of the Microsoft Office Product Group, opens the session at Ignite. She emphasizes that despite the evolution of Office over the years, the core mission remains unchanged: to help users do their best work. The presentation focuses on the latest innovations in Office, specifically new Copilot updates that ensure a consistent and powerful experience whether a user starts in a chat or within a specific app.

The M365 Copilot App Experience13:00

Slide for The M365 Copilot App Experience

The presentation transitions to a demonstration of the M365 Copilot app. This platform serves as a central hub where chat, search, agents, and apps converge. > "We are bringing more of that familiar Office experience directly into chat." The goal is to allow users to stay focused by integrating Office functionalities directly into the chat interface, eliminating the need to constantly switch between different applications.

Streamlined Scheduling with Copilot13:27

Slide for Streamlined Scheduling with Copilot

A live demo showcases how Copilot simplifies administrative tasks like scheduling meetings. - **Context Awareness**: Copilot uses "Work IQ" to identify the correct colleague (e.g., "Nicole") based on past chat and collaboration history. - **Smart Scheduling**: It automatically finds mutually free meeting slots. - **In-Chat Action**: The user can select a time, send the invite, and receive confirmation, all without leaving the chat window.

Office Apps as Agents14:15

Slide for Office Apps as Agents

The concept of integrating Office experiences into chat extends beyond Outlook. Microsoft is transforming core productivity apps—Word, Excel, and PowerPoint—into **agents**. This creates a "chat-first agentic experience," allowing users to leverage the full power of these applications directly within the chat interface to perform complex tasks, such as populating briefing templates or analyzing strategic insights.

Initiating Dashboard Creation14:48

Slide for Initiating Dashboard Creation

The demonstration begins with a user interacting with a chat interface containing business context such as adoption rates and support ticket volumes. The user prompts Copilot to convert this unstructured discussion into a structured **customer dashboard in Excel**. * **Goal**: Create a dashboard to visualize key metrics. * **Input**: Context from the current chat session regarding Contoso.

Context Analysis and Theme Selection15:09

Slide for Context Analysis and Theme Selection

Copilot begins processing the request by analyzing the provided context. It retrieves relevant information from recent files, meetings, and chats. Before generating the final output, Copilot pauses to ask clarifying questions, similar to a human colleague. > "Just to make sure I have clear instructions, please choose some options for your document." The user is presented with **Presentation Options** to select a visual theme for the dashboard.

Reasoning Engine Execution15:34

Slide for Reasoning Engine Execution

Once the theme is selected, the **reasoning model** takes over. This phase involves a multi-step process where Copilot thinks, plans, and executes the task. Key activities displayed in the status window include: * **Creating comprehensive executive dashboard**: Building a production-ready workbook. * **Building Contoso Executive Dashboard**: Structuring sheets with calculations and visualizations. * **Generating trends and insights**: Validating that the output, including formulas and charts, is accurate.

Reviewing the Generated Workbook16:38

Slide for Reviewing the Generated Workbook

The final output is a complete Excel workbook, not just a single table. The workbook includes multiple tabs (e.g., Dashboard, Raw Data) and a summary dashboard titled **"Contoso Executive Dashboard"**. Key features of the generated sheet: * **Visualizations**: Charts showing trends like adoption rates. * **Native Artifacts**: The cells are bound to native Excel formulas, allowing users to click through and audit the calculations.

Refining with Agent Mode17:22

Slide for Refining with Agent Mode

The user opens the generated workbook in the Excel desktop application to make further refinements. They invoke Copilot and switch to **Agent Mode**. * **Functionality**: Agent mode allows users to provide high-level goals rather than specific formula instructions. * **Role**: Copilot acts as a co-editor that can create, understand, and modify the sheet structure based on natural language prompts (e.g., "generate an adoption plan for the next quarter").

Generating Data-Driven Dashboards17:16

Slide for Generating Data-Driven Dashboards

The demonstration begins with Copilot in Excel performing a complex synthesis of data to build an executive dashboard. ### Key Process: - **Deep Reasoning**: The AI executes deep reasoning and complex analysis on the provided dataset. - **Validation**: It validates the data to ensure accuracy before generating the output. - **Output**: The result is a fully populated 'Contoso Executive Dashboard' sheet featuring Key Performance Indicators (KPIs), health scores, and trend charts.

Chain of Thought Transparency17:41

Slide for Chain of Thought Transparency

A crucial feature highlighted is the **Chain of Thought** displayed in the side panel. This feature provides transparency into the AI's logic. > "This makes everybody in the room a power user of Excel." It explains: - How the answer was derived. - Where the formulas are located in the cells. - The specific formulas used. - The relationships between different data points in the sheet.

Native Excel Integration & Model Choice18:11

Slide for Native Excel Integration & Model Choice

The generated content is fully integrated into the native Excel environment rather than being static text or images. - **Live Formulas**: Clicking on a cell reveals standard Excel formulas (e.g., `XLOOKUP`), allowing for manual auditing and editing. - **Native Charts**: Charts are standard Excel objects that can be manipulated using existing chart tools. - **Model Flexibility**: Users can select different underlying models, such as OpenAI or Claude models, to generate the sheet, offering different styles or reasoning capabilities.

Cross-App Workflow in Word19:51

Slide for Cross-App Workflow in Word

The workflow transitions to Microsoft Word to demonstrate **Agent Mode** acting as a real-time writing partner. - **Task**: The user prompts the agent to fill out a 'Customer Internal Briefing' template. - **Data Source**: The agent pulls the synthesized data directly from the previously generated Excel analysis. - **Execution**: Using 'Work IQ', it retrieves the relevant information and populates the Word document, bridging the gap between data analysis and reporting.

Drafting and Refining Documents with Copilot in Word19:44

Slide for Drafting and Refining Documents with Copilot in Word

The demonstration begins with **Copilot in Word** drafting content directly into a document. Key capabilities shown include: - **Contextual Leveraging**: Copilot uses recent files, meetings, chats, and emails to generate relevant content. - **Direct Action**: Instead of just answering questions in a chat pane, Copilot takes action on the canvas, drafting text with proper formatting. - **Multi-step Reasoning**: The user asks Copilot to add an introduction and a follow-up question. The agent intelligently distinguishes between content to add to the document body and questions to answer in the chat interface.

Initiating the PowerPoint Agent20:49

Slide for Initiating the PowerPoint Agent

The workflow transitions to the **Microsoft 365 Copilot app** to create a presentation based on the previous work. > "I want to create a deck now for my internal team presentation." The user selects the **PowerPoint agent** and prompts it to generate an internal briefing deck, instructing it to include adoption data, user feedback, and roadmap highlights derived from the Excel and Word data gathered earlier.

Deep Reasoning and Data Integration21:14

Slide for Deep Reasoning and Data Integration

The PowerPoint agent employs **deep reasoning** and **contextual awareness** to fulfill the request. - **Data Aggregation**: It pulls information from files, emails, reports, and trusted web sources. - **Clarifying Questions**: Similar to a human colleague, the agent pauses to ask clarifying questions to ensure accuracy. In this instance, it asks the user to identify the primary audience (e.g., Executive Leadership) to tailor the presentation's tone and content.

Customization and Deck Generation21:43

Slide for Customization and Deck Generation

The final stage involves customizing the visual style and generating the deck. - **Theme Selection**: The user selects a visual theme (e.g., Bronze) from the provided options. - **Agent Process**: The interface displays the agent's internal process: *Thinking -> Planning -> Reasoning -> Creating -> Validating*. - **Final Output**: The agent produces a polished, multi-slide presentation that synthesizes the provided data into a clear visual story.

Reviewing the Generated Presentation22:13

Slide for Reviewing the Generated Presentation

The speaker reviews the initial output of the PowerPoint deck generated by Copilot. Key observations include: * **Clean Layout**: The slides are organized professionally. * **Cohesive Design**: The visual style is consistent throughout. * **Compelling Story**: The narrative is aligned with the user's insights.

Editing Native PowerPoint Artifacts22:29

Slide for Editing Native PowerPoint Artifacts

The presentation is opened in the PowerPoint desktop application to demonstrate editability. > "All of these artifacts... are native PowerPoint artifacts." The speaker emphasizes that visuals, charts, graphs, and images are fully editable native elements, allowing users to refine the deck using standard PowerPoint tools.

Introduction to Copilot Notebooks23:28

Slide for Introduction to Copilot Notebooks

The speaker introduces **Copilot Notebooks**, a workspace designed for managing shared projects. * **Purpose**: To bring together all project artifacts such as files, documents, meetings, chats, and emails. * **Context Scoping**: When a user asks Copilot a question within a Notebook, the search is scoped specifically to the reference documents in that Notebook, rather than the entire data graph.

Adding References to Notebooks24:10

Slide for Adding References to Notebooks

A demonstration of adding the newly created PowerPoint deck to the Notebook. 1. The user selects "Add references." 2. Copilot locates the recent deck. 3. Once added, Copilot generates an **overview page** with a summary of the document's content and key insights.

Audio Overview Feature24:30

Slide for Audio Overview Feature

The speaker showcases the **Audio Overview** feature within the Notebook. * This feature creates an engaging audio summary of the entire notebook's content. * It allows users to listen to a synthesis of their project documents, similar to a podcast, while on the go.

Microsoft 365 Copilot Agentic Capabilities24:49

Slide for Microsoft 365 Copilot Agentic Capabilities

The presentation highlights the new agentic capabilities of Microsoft 365 Copilot, emphasizing the transition from information overload to actionable knowledge. ### Key Benefits * **Idea to Impact:** Facilitates a smoother workflow from conception to execution. * **True Collaborator:** Copilot is positioned as a partner that helps users: * Create * Think * Execute The slide lists specific features such as scheduling meetings in chat, agent modes in Excel, Word, and PowerPoint, and the creation of specific agents for these applications.

Agent Mode in Microsoft Edge25:27

Slide for Agent Mode in Microsoft Edge

Nicole introduces **Agent Mode** for the Microsoft Edge browser, designed to make the browser smarter by integrating search, chat, and navigation. > "Instead of just searching, Copilot now takes on tasks across your websites." ### Functionality * **Task Automation:** Copilot can perform actions like booking travel or submitting timesheets directly within the browser context. * **Demo:** The video demonstrates Copilot automatically filling out a weekly timesheet based on user prompts.

Pattern 2: AI Teammates25:44

Slide for Pattern 2: AI Teammates

The lecture introduces the second of three patterns of work: **AI Teammates**. This concept shifts the role of AI from a passive tool to an active participant in collaboration. ### The Shift in Work * **Collaboration:** Acknowledging that the best work happens in teams where ideas spark and problems are solved. * **Active Contribution:** Imagining a future where AI teammates actively contribute to meetings and projects rather than just observing.

Product Launch Readiness Demo27:08

Slide for Product Launch Readiness Demo

A demonstration of an AI teammate workflow involving product management tasks. The presenter uses Copilot to review a product roadmap before a client meeting. ### Workflow Steps 1. **Status Check:** The user prompts Copilot to generate a view of the "ZavaCore Product Launch Readiness." 2. **Analysis:** The system generates a detailed readiness table, identifying components that are "On Track" (green) and those "At Risk" (red). 3. **Actionable Collaboration:** Instead of manually copying data to an email, the user utilizes a **Teams mode** feature to instantly start a group chat with the product lead directly from the Copilot interface to address the risks.

Transitioning from Individual to Group Collaboration27:08

Slide for Transitioning from Individual to Group Collaboration

The demonstration begins by moving a one-on-one chat between a user and an AI agent into a **Microsoft Teams** group chat. This transition shifts the context from individual work to a collaborative environment. * **Context Switching:** The chat history and context are preserved as the conversation moves to Teams. * **Adding Collaborators:** A colleague, Jeff Teper, is added to the chat to address specific product issues identified by the AI.

Multi-User Collaboration with Copilot27:47

Slide for Multi-User Collaboration with Copilot

Jeff Teper takes the stage to demonstrate the receiver's perspective in Teams. The interface shows the shared chat history, including the context handed over from the previous one-on-one session. This illustrates seamless continuity in multi-user collaboration scenarios involving AI.

Querying Copilot for Shared Insights27:59

Slide for Querying Copilot for Shared Insights

In the group chat, Jeff engages Copilot to assist with the discussion. He prompts the AI to identify the biggest product learnings based on recent customer meetings. > "Copilot, capture the 3 biggest product learnings based on recent customer meetings..."

Privacy and Data Security in Group Chats28:17

Slide for Privacy and Data Security in Group Chats

While Copilot processes the request, a critical security feature of **Teams mode** is explained: * **Data Boundaries:** Copilot does not access individual user memories (private data of either participant). * **Access Control:** It ensures that all content used to generate the response is accessible to *both* users in the chat. * **Verification:** As a final safety step, Copilot requires explicit permission from the user to post the generated response back to the group chat.

Reviewing and Sharing AI Responses29:05

Slide for Reviewing and Sharing AI Responses

Copilot returns a draft response visible only to the requester initially. 1. **Review:** The AI provides three distinct learnings with specific customer examples. 2. **Approval:** The user reviews the accuracy and relevance of the information. 3. **Sharing:** By clicking **Allow**, the response is posted to the shared chat stream, making it visible to all participants.

Ad-hoc vs. Long-term Collaboration29:20

Slide for Ad-hoc vs. Long-term Collaboration

The lecture concludes by distinguishing between different collaboration modalities in Teams: * **Group Chats:** Ideal for ad-hoc, quick, and secure conversations to resolve immediate issues. * **Channels:** Better suited for long-term projects and structured collaboration, distinct from the transient nature of chats.

Channel Agents and Jira Integration29:36

Slide for Channel Agents and Jira Integration

## Dedicated Channel Agents Microsoft Teams channels now feature dedicated agents equipped with specific knowledge and skills tailored to the channel's activity. In this demonstration: * The user interacts with the **Fiber Launch Agent**. * A request is made to pull the latest issues directly from **Jira**. * This highlights the agent's ability to collaborate durably within the team's workspace without needing to switch contexts.

Reasoning with Model Context Protocol (MCP)30:03

Slide for Reasoning with Model Context Protocol (MCP)

## How the Agent Works The agent processes the request using its reasoning capabilities and the **Model Context Protocol (MCP)**. > "Channels have a dedicated agent that has the specific knowledge and skills that are associated with the activity in the channel." **Key Technical Details:** * **MCP Support:** Atlassian Jira supports MCP, allowing the agent to understand and utilize Jira's tools. * **Reasoning Engine:** The agent identifies the necessary tool, calls it to retrieve data, and formulates a response. * **Result:** The agent returns a list of four specific issues from Jira directly into the chat thread.

Microsoft 365 Skills: Scheduling Meetings30:51

Slide for Microsoft 365 Skills: Scheduling Meetings

## Cross-App Functionality Channel agents can also leverage skills across the Microsoft 365 ecosystem. They can perform actions such as making documents, creating tasks, and scheduling meetings. **Workflow Demonstrated:** 1. The user identifies a need for a meeting based on the Jira issues. 2. The agent is asked to schedule a meeting between the user and a colleague (Robin). 3. The agent analyzes both users' calendars to find the next available common time slot. 4. The agent proposes a time, and the user confirms the booking with a single click.

The Meeting Facilitator Agent31:30

Slide for The Meeting Facilitator Agent

## In-Meeting Assistance The demonstration transitions to a live meeting environment to showcase the **Facilitator** agent, a special-purpose agent designed for meetings. **Features Shown:** * **Real-time Agenda Management:** The user notices the agenda is incomplete and asks the Facilitator to add "Success metrics." * **UI Update:** A new bar at the top of the Teams meeting window displays the live agenda, ensuring all participants are aware of the meeting's structure and progress.

AI-Powered Meeting Facilitator32:04

Slide for AI-Powered Meeting Facilitator

### Keeping Meetings on Track The speaker introduces a new AI-powered capability in Microsoft Teams called **Facilitator**. This feature is designed to solve the common problem of running out of time in meetings before covering all agenda items. * **Real-time Monitoring**: The Facilitator monitors the conversation and tracks progress against the agenda. * **Visual Progress**: It displays which topics have been covered (e.g., content creation) and how much time remains for subsequent topics (e.g., success metrics). * **User Experience**: The feature aims to keep discussions focused and provides a satisfying conclusion to well-managed meetings.

Real-time Document Generation32:39

Slide for Real-time Document Generation

### Automating Work During Meetings The Facilitator goes beyond tracking time; it can actively perform tasks while the meeting is in progress. > "The facilitator can get work done while the meeting's going on." In the demonstration, a user asks the AI to prepare a strategy document based on the current discussion. The system successfully generates a structured Word document, complete with product positioning and key features, allowing participants to focus on the conversation rather than manual drafting.

Live Notes and Task Integration33:52

Slide for Live Notes and Task Integration

### Automated Notes and Follow-up One of the highlighted features is **Live Meeting Notes**, which automatically captures and organizes the discussion. * **Auto-Summarization**: Notes are typed, updated, and categorized in real-time as the meeting progresses. * **Planner Integration**: The system tracks follow-up tasks using Microsoft Planner. The AI can suggest tasks, and users can manually add, reassign, or modify them. This integration ensures that meetings result in effective outcomes with clear accountability.

SharePoint Knowledge Management35:22

Slide for SharePoint Knowledge Management

### Transition to Knowledge Management The presentation shifts focus from collaboration in Teams to knowledge management in **SharePoint**. * **Work IQ**: The speaker references the concept of "Work IQ," where the collective data of an organization powers AI capabilities like Copilot. * **Scale**: SharePoint is described as the repository for billions of new documents added daily, serving as a critical foundation for organizational intelligence through its compliance and collaboration features.

SharePoint Knowledge Agent34:41

Slide for SharePoint Knowledge Agent

The speaker introduces a product catalog in SharePoint and activates the **Knowledge Agent**. He selects the "Organize this library" option to demonstrate how AI can assist in managing document libraries. He instructs the agent to automatically create specific columns for the library: - Product Name - Category - Material

Automated Metadata Extraction35:27

Slide for Automated Metadata Extraction

The AI processes the documents and populates the requested metadata columns. The speaker highlights a common issue: while metadata is essential for organizing, filtering, and workflows, users often dislike manually entering it. The AI overcomes this by extracting complex details, such as material composition, directly from the documents.

Copilot Data Query36:37

Slide for Copilot Data Query

The demonstration moves to **Copilot** to illustrate how it utilizes the enriched metadata. The speaker issues a specific prompt: > "Create a table of only the ZavaCore apparel in stock in Florida." Copilot successfully uses the inferred metadata (location, stock status, product type) to generate a precise table answering the user's request.

ChatGPT vs. Copilot Comparison37:47

Slide for ChatGPT vs. Copilot Comparison

The speaker compares the results with **ChatGPT** using SharePoint connectors. When asked the exact same question, ChatGPT fails to retrieve the correct data or generate the table. It struggles to find the information, demonstrating that unlike Copilot in Microsoft 365, it does not leverage the deep grounding and enriched data available in the SharePoint environment.

Summary of Copilot in Teams and SharePoint37:21

Slide for Summary of Copilot in Teams and SharePoint

The presentation concludes the section on Copilot's integration into collaboration and knowledge management tools. ### Key Takeaways * **Teams:** Copilot assists with collaboration through features like the **Channel agent**. * **SharePoint:** Focuses on knowledge management with the **Knowledge agent**. * **New Announcements:** Several new capabilities were announced to build intelligence into SharePoint and Work IQ.

Pattern 3: AI as Operators37:39

Slide for Pattern 3: AI as Operators

Nicole introduces the third pattern of work: **AI as operators**. This pattern represents a shift in how AI is utilized within business processes. > "More and more, humans will set goals and handle exceptions, and then agents will execute tasks." ### Scope of AI Operators * **Improving Processes:** Enhancing existing business workflows. * **Reimagining Business:** Fully transforming business operations with AI.

Workforce Insights Agent Demo38:42

Slide for Workforce Insights Agent Demo

A demonstration of the **Workforce Insights agent** addresses the challenge of understanding workforce capacity, specifically for upcoming events like product launches. ### Agent Capabilities * Provides leaders with a clear view of their workforce. * Offers guidance on hiring and skilling. * Assists with talent mobility and allocation.

Querying Organizational Staffing39:10

Slide for Querying Organizational Staffing

The presenter interacts with the agent using a natural language prompt to analyze the organization's readiness for a product launch. **The Prompt Requests:** * An understanding of current staffing. * A breakdown of skills. * Identification of gaps based on company benchmarks. *Note: The agent uses a reasoning model to process this complex request.*

Analysis Results: Skills Overview39:38

Slide for Analysis Results: Skills Overview

The agent generates a comprehensive report. The first section, **"Your org at a glance,"** visualizes the distribution of skills across the entire organization, providing a high-level baseline of available talent.

Analysis Results: Critical Skills Coverage40:25

Slide for Analysis Results: Critical Skills Coverage

The report drills down into **"Launch-critical skills coverage."** ### Visualization Features * **Bar Chart:** Displays coverage for specific skills required for the launch. * **Status Indicators:** Categorizes skill levels as Strong, Moderate, or Weak. * **Gap Identification:** Highlights areas where the organization is low on key skills necessary for a successful launch.

Analysis Results: Targeted Recommendations40:37

Slide for Analysis Results: Targeted Recommendations

The final section of the report provides **"Targeted recommendations"** on where to add support. ### Actionable Insights * **Specific Guidance:** Tells leaders exactly how to handle medium and high-priority areas. * **Resource Allocation:** Suggests designating specific roles or adding headcount to address bottlenecks. * **Detailed Steps:** Moves beyond data visualization into a recommendation mode for decision-making.

Strategic Hiring Decision39:28

Slide for Strategic Hiring Decision

The speaker concludes a strategic analysis regarding resource allocation. Faced with the choice of upskilling existing staff or acquiring new talent to ensure a strong product launch, the decision is made to **hire a new marketing resource**.

The Challenge of Onboarding39:43

Slide for The Challenge of Onboarding

With a new team member hired, the focus shifts to the **onboarding process**, which is described as painful and logistics-heavy. Key pain points include: * Procuring hardware (laptops) * Managing distribution lists * Handling administrative tasks that distract from core job responsibilities

Introduction to Project Opal40:11

Slide for Introduction to Project Opal

### Project Opal Overview **Project Opal** is introduced as an agent designed to execute task-based work. Its primary goal is to remove "drudgery work" from the user's workload. > "This is an early stage product, but it's a great example of where work is going with the support of agents."

Initiating an Onboarding Job41:12

Slide for Initiating an Onboarding Job

The demonstration begins within the Opal interface. The user initiates a new job to onboard a specific employee (Rachel). * **Input:** The user describes the work to be done. * **Process:** Opal begins the workflow, which typically takes time but is accelerated for the demonstration.

Opal's Execution Environment41:32

Slide for Opal's Execution Environment

Opal operates by spinning up a **Windows 365 Cloud PC**. * **Access:** This allows the agent to access all organizational apps and sites via a browser. * **Technology:** It utilizes a reasoning model combined with computer vision to navigate interfaces and perform tasks autonomously.

Automating Procurement41:53

Slide for Automating Procurement

The agent navigates to the company's **PC Request Portal**. Having parsed the onboarding checklist, Opal automatically fills out the necessary forms to configure and request a laptop for the new hire.

Human-in-the-Loop Guardrails42:51

Slide for Human-in-the-Loop Guardrails

The system demonstrates built-in **guardrails**. When Opal encounters a high-stakes action, such as spending money on hardware, it pauses for approval. 1. **Notification:** Opal alerts the user that attention is needed. 2. **Intervention:** The user clicks "Take Control" to interact directly with the Cloud PC session. 3. **Approval:** The user submits the request manually. 4. **Handoff:** Control is returned to Opal to continue the workflow.

Resuming Automated Tasks43:06

Slide for Resuming Automated Tasks

After the user intervention, Opal returns to its task list. It proceeds to the next step in the onboarding process, which involves navigating to security settings to add the employee to the appropriate **security groups**.

Project Opal Demo: Automating Onboarding41:56

Slide for Project Opal Demo: Automating Onboarding

### Automating Multi-Step Tasks The video begins with a demonstration of **Project Opal**, an advanced agent designed to handle complex workflows. In this scenario, the agent is onboarding a new marketing hire. * **Automated Actions:** The agent navigates to the security group manager, identifies the correct security group, and adds the new hire to specific distribution lists required for her role. * **User Experience:** The interface shows the agent actively typing and filling out justifications, described by the presenter as "mesmerizing." The user simply observes as the work is completed. * **Review & Completion:** Once the activities are finished, the agent prompts the user for a final review. The user confirms the job is done, and the task is marked as complete within the system.

Redefining Business Processes with Agents42:40

Slide for Redefining Business Processes with Agents

### The Power of Autonomous Agents The presenter returns to the stage to discuss the broader implications of agents like Opal. These agents are capable of executing multi-step processes with minimal human intervention. > "At Microsoft, we're using Opal for all kinds of tedious tasks such as compliance audits and invoice processing." **Key Benefits:** * **Redefining Process:** Agents help reimagine workflows from workforce planning to full execution. * **Efficiency:** They handle routine tasks, freeing up human employees for higher-value work. The segment concludes with an introduction to a customer showcase featuring **Bristol Myers Squibb**.

Bristol Myers Squibb: AI for Employee Experience43:20

Slide for Bristol Myers Squibb: AI for Employee Experience

### Enhancing Productivity at BMS **Mike DiPaolo**, Senior Director of Productivity Ecosystem Services at Bristol Myers Squibb, takes the stage to share their journey with AI agents. * **Objective:** BMS aims for employees to experience and "truly fall in love" with AI by simplifying their daily work. * **Strategy:** They partnered with IT and People Services to develop agents using **Agent Builder**. * **Target Use Case:** The focus is on automating essential but often disliked tasks, specifically **performance reviews**.

Demo: The Self-Reflection Agent43:49

Slide for Demo: The Self-Reflection Agent

### Automating Performance Reviews Mike demonstrates the **MySelfReflection** agent, designed to assist employees in writing their self-evaluations. **How it Works:** 1. **Input:** The user provides their job title and grade level via an engineered prompt. 2. **Data Synthesis:** The agent scans the user's emails, chats, documents, and project files. 3. **Contextual Knowledge:** It cross-references this data with a SharePoint site containing grade-level specific performance guidelines. 4. **Output:** The agent generates a detailed draft of the self-reflection, categorizing achievements under organizational goals and core values, which the employee can then review and refine.

Automating Employee Performance Reviews44:24

Slide for Automating Employee Performance Reviews

To address the challenge of managers handling more direct reports, a specialized agent has been created to assist with performance reviews. - **Engineered Prompting**: Managers simply input the employee's name into a pre-configured prompt. - **Data Scanning**: The agent scans various data sources, including conversations, project work, SharePoint sites, chats, and emails to gather relevant performance data.

Generating Data-Driven Summaries44:49

Slide for Generating Data-Driven Summaries

The agent processes the gathered data to produce a comprehensive performance report. > "I get an overall performance summary of what she's done over the year. Even if she hasn't directly reported to me the entire year." **Key Output Sections:** - Overall Performance Summary - Key Achievements - Feedback Received - Development Areas This automation returns time to managers, allowing them to focus on making the reviews meaningful and actionable rather than spending hours on data collection.

The Manager Hub and Work Transformation45:29

Slide for The Manager Hub and Work Transformation

Beyond performance reviews, a **Manager Hub** is being developed to streamline administrative tasks. - **Centralized Queries**: Allows querying policies, PTO time, and training documents from a single agent interface without logging into multiple systems. - **Strategic Goal**: The initiative aims to transform workflows through a partnership with Microsoft, focusing on making AI tools meaningful for individual employees rather than just the business entity.

Specialized Pilot Agents46:46

Slide for Specialized Pilot Agents

Several other agents are currently in pilot to support diverse workforce needs: - **Neurodiversity Support Agent**: Designed to level the playing field and give employees a voice. - **Meeting Preparation Assistant**: Facilitates the creation of pre-reads and meeting documents. - **IT Contract Reviewer Agent**: Developed with the tuning team to handle regulatory and legal review responsibilities. - **TPI Tracker Agent**: (Displayed on slide as part of the agent suite).

Three Patterns of Work46:52

Slide for Three Patterns of Work

Copilot and agents are transforming the workplace through three distinct patterns: - **AI Assistants** - **AI Teammates** - **AI Operators**

AI Built for Work47:02

Slide for AI Built for Work

The core differentiator of Copilot is that it is not generic AI, but **AI Built for Work**. Key characteristics include: - **Real-time Data:** Built with your data and full context. - **Integration:** Embedded in the tools employees use every day. - **Security:** Built securely and remains under user control. This approach is designed for real work, real outcomes, and real responsibility.

Closing Remarks47:33

Slide for Closing Remarks

The presentation concludes with thanks to the customers for their trust and feedback. The speaker emphasizes that customer achievement is their primary inspiration and transitions to a customer story from **PwC**.

PwC Case Study: Scale and Complexity47:55

Slide for PwC Case Study: Scale and Complexity

PwC shares their experience adopting AI, highlighting the challenge of sheer scale: - Operating in **130+ countries**. - Managing **500,000 mailboxes**. - Migrating **15 petabytes** of data. They navigated complex regulations and requirements for every country while ensuring no business disruption.

Partnership and Deployment48:26

Slide for Partnership and Deployment

Through a partnership with Microsoft, PwC successfully deployed the technology to give their workforce better tools. - **Global Enablement:** 210,000 people are enabled on Copilot today. - **Embedded Workflow:** The AI is embedded in daily tools, acting as a teammate that understands the user's specific work context.

The Agentic Era48:46

Slide for The Agentic Era

In the **Agentic Era**, the focus shifts to scale, robustness, and adoption. > "I want to make sure that our users don't have to think about things like compliance or security." PwC is building agents that function like team members, such as an agent specifically designed to help draft user stories, allowing for faster turnaround times.

Rapid Adoption of Copilot49:20

Slide for Rapid Adoption of Copilot

### Skyrocketing Usage The video begins by highlighting the massive scale of adoption for Microsoft Copilot within the organization. - **Key Metric**: Over **40 million Copilot actions** have been recorded in the past six months. - **Trend**: Adoption rates are described as "skyrocketing," indicating immediate and widespread utility among users.

Seamless Integration and Model Agnosticism49:33

Slide for Seamless Integration and Model Agnosticism

### A Model Agnostic Approach Speakers discuss the technical and operational benefits of their implementation strategy. > "This model agnostic approach enables us continually to leverage Copilot and know that we're not going to be locked out of features." **Key Benefits:** - **Feature Access**: Ensuring continuous access to the latest capabilities. - **Workflow Continuity**: Eliminating the need to constantly switch between different tools or solutions to get work done.

Investing in Scalable Partnerships49:42

Slide for Investing in Scalable Partnerships

### Choosing the Right Technology Partner The narrative shifts to the strategic importance of vendor selection in a rapidly evolving AI landscape. - **Strategic Alignment**: It is crucial to pick partners, such as Microsoft, that are investing at the necessary scale. - **Reliability**: Relying on a partner capable of supporting the massive volume of actions seen in enterprise environments.

Limitless Future Operations49:50

Slide for Limitless Future Operations

### The 12-Month Outlook The speakers project a transformative future for day-to-day business operations. - **Limitless Potential**: Within 12 months, the possibilities for using Copilot to activate and support operations are viewed as nearly limitless. - **Operational Support**: The focus is on integrating AI deeply into the daily functions of the firm. - **Conclusion**: The future of this technology integration is described as "exciting."

Session Feedback50:03

Slide for Session Feedback

### Event Survey A call to action for attendees to provide feedback on the session. - **Action**: Scan the QR code or visit the provided URL. - **Purpose**: To share thoughts on the sessions and complete the event survey.

Closing50:06

Slide for Closing

### Conclusion A final closing slide thanking the audience for joining the session.